Circadian rhythms and dementia: sensors remain far from a validated predictive tool

Measuring circadian rhythm disruptions does not mean being able to predict dementia. A systematic review of older adults or people with cognitive vulnerability reports associations with functioning and cognitive performance. But none of the artificial intelligence models identified had been validated using external data: their use as predictive biomarkers remains exploratory.

A wrist-worn sensor tracking an older person’s activity rhythms.
Gadgets & TechEditorial illustration, not study data.

Ten studies included, with very different populations

The review brings together ten eligible studies published between 2015 and 2025, plus three sensitivity references used to define the scope of the evidence base. Study sizes ranged from 16 to 7,297 participants. Populations included both community-dwelling older adults and care home residents with advanced dementia.

Wrist actigraphy was the predominant monitoring method. The parameters studied included day-to-day rhythm stability, fragmentation throughout the day, amplitude and peak timing. The heterogeneity of the outcomes measured prevented a meta-analysis: the authors provide a narrative synthesis.

An observed association, not demonstrated prediction

Circadian rhythm fragmentation was associated with functional decline, dementia risk and cognitive performance. However, these observations do not demonstrate that rhythm disruptions cause neurodegeneration. They may also be a consequence of it.

The authors rate the certainty of the evidence as moderate for associations between actigraphy-measured fragmentation and functioning among people living in care facilities. It is low for claims of predictive value among community-dwelling people. These two conclusions are therefore not interchangeable.

The “AI” label masks a range of approaches

The methods identified included supervised learning, unsupervised clustering, and algorithm development or validation. However, seven studies used deterministic algorithms without fitting a predictive model. Not all the studies were therefore testing AI capable of estimating future risk.

No AI model had undergone external validation, and no study followed AI-specific reporting guidelines. Most studies were descriptive or observational, with samples that were often small and lacked diversity. These limitations mean the findings cannot be considered clinical validation of the devices.

What this review changes

Sensors offer a potential way to document circadian disruptions in everyday life. This review is not enough to justify buying a device to predict Alzheimer’s disease or another neurodegenerative disease.

The authors call for longitudinal validation across multiple sites, using standardized measures and diverse populations, before clinical deployment as neurodegenerative biomarkers.

Limitations to keep in mind

Only the PubMed abstract was consulted, not the full text. The studies are heterogeneous, mostly descriptive or observational, and no AI model has been validated using external data. Funding and conflicts of interest were not checked.

What we can take away

Sensors can document disrupted rhythms. Their ability to predict neurodegenerative disease in an individual remains to be validated.

Sources and provenance

Partager cet article

X f in

An error or a source to report? Contact the editorial team.